Every PE firm sits on a goldmine it cannot access.

Decades of deal flow. Hundreds of sector deep-dives. Thousands of relationships cultivated across management teams, intermediaries, and co-investors. The reasoning behind every deal pursued, and every deal passed on.

This is a firm’s institutional knowledge, its most proprietary and defensible asset. Today, the vast majority of it is dark data: scattered across disconnected systems, trapped in individual memories, buried in file storage no one searches.

No foundation model, no matter how capable, can access what it has never seen. And no AI tool, no matter how fast, can replicate the judgment a firm has built over decades if that judgment was never captured in the first place.

Private Markets: The Conviction Equation

PE is a conviction business. The ultimate measure is not how fast you process information. It is how quickly you can build conviction on the opportunities that matter.

Building conviction is an iterative and compounding process. Each stage of a deal generates knowledge that should inform the next, and each deal's knowledge should compound into the next. 

Today, that compounding rarely happens. The four critical inputs to conviction are structurally constrained: data, time, perspective, and iteration.

The natural instinct is to throw AI at the problem. But the question is: which layer of AI actually addresses these constraints?

When Generic-AI Becomes Table Stakes

The arrival of powerful general-purpose AI is real and meaningful. Models like Claude and GPT can summarize a CIM in seconds, draft an IC memo, and answer questions across uploaded documents.

But here is the question that should keep every PE firm honest: when every firm in the market can access the same model, where is the edge?

Summarizing a CIM faster does not help you see what others miss. These are efficiency gains at the task level. Valuable, but not differentiating. 

The firms that will win the next decade will be those that figure out how to activate what no competitor can replicate: their own institutional knowledge.

Three Layers of Institutional Intelligence

Not all intelligence is created equal. To understand why most AI tools plateau at productivity and never reach conviction, it helps to see the full picture. Institutional intelligence operates in three layers, each building on the one below.

Level 1: Document Intelligence

This is where most AI tools operate today. Read a file, run analysis, generate a report. Useful, but stateless. Every interaction starts from zero, with no memory of what the firm has seen before.

Connecting these tools to more data sources through integrations helps, but it doesn't change what's underneath. Every conversation still starts from a blank slate, and the firm's knowledge never builds on itself.

Level 2: Entity Intelligence

This layer goes beyond documents to understand the relationships between things such as companies, people, deals, sectors, and deal activities. It surfaces connections that no individual team member could hold in their head: 

This target’s supply chain mirrors a deal we diligenced two years ago.

What makes this possible is a knowledge graph: a persistent data structure that maps every entity your firm has ever touched and the relationships between them. Unlike a file system or a CRM, it doesn't just store information. It connects it. 

This is a fundamentally different architecture from task-level AI, and we'll go deeper on how it works in our next piece.

Level 3: Conviction Intelligence

This is where institutional knowledge becomes an institutional edge. Conviction Intelligence is what happens when institutional knowledge flows across every stage of the deal lifecycle as a compounding system. Not by replacing human judgment, but by ensuring that judgment is informed by everything the firm collectively knows.

Consider what this looks like in practice:

A deal team is evaluating a healthcare services target. The system cross-references the opportunity against the firm’s full history and surfaces three patterns the team had not considered: 

1. Revenue mix closely mirrors a business the firm acquired four years ago that outperformed its underwriting by 30%.

2. The management team’s background overlaps with a portfolio company’s board, suggesting a relationship angle no one had mapped.

3. A similar business in the same sub-sector was passed on two years ago for operational concerns that ultimately impacted margins. 

One of these insights reshapes the pricing thesis. Another accelerates the relationship-building timeline by weeks. None of them would have surfaced through document summarization alone. 

Most tools in the market today operate at Level 1. A few attempt pieces of Level 2. Level 3 requires something that cannot be retrofitted onto a general-purpose AI. It requires a purpose-built infrastructure. One designed to compound intelligence with every deal, not reset with every conversation. That's what we set out to build.

The Asset That Compounds

The AI landscape will keep evolving. Models will get faster, cheaper, more capable. Today’s tools will be replaced by tomorrow’s. This is the nature of the application layer. 

What does not commoditize is the structured, proprietary knowledge infrastructure that lives underneath—the institutional knowledge that makes AI yours. 

This is the asset most PE firms have not yet built. And unlike a software subscription you can adopt overnight, institutional intelligence has a time dimension: the earlier you start capturing and structuring your firm’s knowledge, the richer the graph becomes, the smarter every workflow gets, and the wider the gap becomes between you and competitors that start later.

Models are interchangeable. Your institutional knowledge is not.

See what this looks like for your firm